timetk: A toolkit for time series analysis in the tidyverse

This tutorial focuses on, plot_time_series(), a workhorse time-series plotting function that:

Generates interactive plotly plots (great for exploring & shiny apps)
Consolidates 20+ lines of ggplot2 & plotly code
Scales well to many time series
Can be converted from interactive plotly to static ggplot2 plots

library(tidyverse)
library(lubridate)
library(timetk)

Plotting Single Time Series

Let’s start with a popular time series, taylor_30_min, which includes energy demand in megawatts at a sampling interval of 30-minutes. This is a single time series.

taylor_30_min
#> # A tibble: 4,032 x 2
#>    date                value
#>    <dttm>              <dbl>
#>  1 2000-06-05 00:00:00 22262
#>  2 2000-06-05 00:30:00 21756
#>  3 2000-06-05 01:00:00 22247
#>  4 2000-06-05 01:30:00 22759
#>  5 2000-06-05 02:00:00 22549
#>  6 2000-06-05 02:30:00 22313
#>  7 2000-06-05 03:00:00 22128
#>  8 2000-06-05 03:30:00 21860
#>  9 2000-06-05 04:00:00 21751
#> 10 2000-06-05 04:30:00 21336
#> # … with 4,022 more rows

The plot_time_series() function generates an interactive plotly chart by default.

Simply provide the date variable (time-based column, .date_var) and the numeric variable (.value) that changes over time as the first 2 arguments
.plotly_slider = TRUE adds a date slider to the bottom of the chart.

taylor_30_min %>%
  plot_time_series(date, value, .plotly_slider = TRUE)

Plotting Groups

Next, let’s move on to a dataset with time series groups, m4_daily, which is a sample of 4 time series from the M4 competition that are sampled at a daily frequency.

m4_daily %>% group_by(id)
#> # A tibble: 9,743 x 3
#> # Groups:   id [4]
#>    id    date       value
#>    <fct> <date>     <dbl>
#>  1 D10   2014-07-03 2076.
#>  2 D10   2014-07-04 2073.
#>  3 D10   2014-07-05 2049.
#>  4 D10   2014-07-06 2049.
#>  5 D10   2014-07-07 2006.
#>  6 D10   2014-07-08 2018.
#>  7 D10   2014-07-09 2019.
#>  8 D10   2014-07-10 2007.
#>  9 D10   2014-07-11 2010 
#> 10 D10   2014-07-12 2002.
#> # … with 9,733 more rows

Visualizing grouped data is as simple as grouping the data set with group_by() prior to piping into the plot_time_series() function. Key points:

Groups can be added in 2 ways: by group_by() or by using the ... to add groups.
Groups are then converted to facets.
.facet_ncol = 2 returns a 2-column faceted plot
.facet_scales = "free" allows the x and y-axis of each plot to scale independently of the other plots

m4_daily %>%
  group_by(id) %>%
  plot_time_series(date, value,
                   .facet_ncol = 2, .facet_scales = "free")

Adjusting the Smoother

By default, a LOESS smoother is added to capture trend. This smoother is similar to geom_smooth(). However, there are advanced controls including being able to change the flexibility of the smoother using time-based phrases:

.smooth_period = "auto" Selects 75% of the data for the smoother span.
.smooth_period = "2 years" Each series uses a loess trend consisting of 2 full years of data.
.smooth_period = 365.5 * 2 Selects exactly 731 observations.
.smooth_span = 0.75 Each series uses 75% of the data. Overrides the .smooth_period.

Note that setting smooth_message = TRUE returns a message with the number of observations used for the loess trend.

m4_daily %>%
  group_by(id) %>%
  plot_time_series(date, value,
                   .facet_ncol = 2, .facet_scales = "free",
                    # Note that D10 and D410 have less than 730 observations
                   .smooth_period  = "2 year",
                   # Shows the trend period selected
                   .smooth_message = TRUE)
#> trend = 673 days
#> trend = 731 days
#> trend = 675 days
#> trend = 730 days

Visualizing Transformations & Sub-Groups

Let’s switch to an hourly dataset with multiple groups. We can showcase:

Log transformation to the .value
Use of .color_var to highlight sub-groups.

m4_hourly %>% group_by(id)
#> # A tibble: 3,060 x 3
#> # Groups:   id [4]
#>    id    date                value
#>    <fct> <dttm>              <dbl>
#>  1 H10   2015-07-01 12:00:00   513
#>  2 H10   2015-07-01 13:00:00   512
#>  3 H10   2015-07-01 14:00:00   506
#>  4 H10   2015-07-01 15:00:00   500
#>  5 H10   2015-07-01 16:00:00   490
#>  6 H10   2015-07-01 17:00:00   484
#>  7 H10   2015-07-01 18:00:00   467
#>  8 H10   2015-07-01 19:00:00   446
#>  9 H10   2015-07-01 20:00:00   434
#> 10 H10   2015-07-01 21:00:00   422
#> # … with 3,050 more rows

The intent is to showcase the groups in faceted plots, but to highlight weekly windows (sub-groups) within the data while simultaneously doing a log() transformation to the value. This is simple to do:

.value = log(value) Applies the Log Transformation
The data is ungrouped, so we can add facets internally using the ... to supply one or more facet columns.
.color_var = week(date) The date column is transformed to a lubridate::week() number. The color is applied to each of the week numbers.

m4_hourly %>%
  plot_time_series(date, log(value),             # Apply a Log Transformation
                   id,                           # Faceting (Group) 
                   .color_var = week(date),      # Color applied to Week transformation
                   # Facet formatting
                   .facet_ncol = 2,
                   .facet_scales = "free")

Static ggplot2 Visualizations & Customizations

All of the visualizations can be converted from interactive plotly (great for exploring and shiny apps) to static ggplot2 visualizations (great for reports).

taylor_30_min %>%
  plot_time_series(date, value,
                   .color_var = month(date, label = TRUE),

                   # Returns static ggplot
                   .interactive = FALSE,

                   # Customization
                   .title = "Taylor's MegaWatt Data",
                   .x_lab = "Date (30-min intervals)",
                   .y_lab = "Energy Demand (MW)",
                   .color_lab = "Month") +
  scale_y_continuous(labels = scales::comma_format())

Learning More

If you are interested in learning from my advanced Time Series Analysis & Forecasting Course, then join my waitlist. The course is coming soon.

You will learn:

Time Series Preprocessing, Noise Reduction, & Anomaly Detection
Feature engineering using lagged variables & external regressors
Hyperparameter Tuning
Time series cross-validation
Ensembling Multiple Machine Learning & Univariate Modeling Techniques (Competition Winner)
NEW - Deep Learning with RNNs (Competition Winner)
and more.

Signup for the Time Series Course waitlist